Early Detection of Diabetic Retinopathy Using Artificial Intelligence: Impact on Physical Performance and Psychosocial Outcomes

Abstract

Moamen Abdelfadil Ismail*, Warif Nasser Alghofaily, Najla Abdelhadi Abdalla, Ghazi Awad A Al Qahtani, Talal Abdulmalek Almalki, Amani Abdulmanam Alali, Abdullah Eid A Alsobaie, Asayil Ahmed Alrasheed, Omama Abubaker Alamin, Batool Mohammed alhashidi, Taha Isam Khayat

Background: Diabetic retinopathy (DR) remains a leading cause of preventable vision impairment globally. Traditional screening programs are resource-intensive and underutilized, especially in low- and middle-income regions. The integration of artificial intelligence (AI) into ophthalmology presents a promising solution to enhance early detection and diagnosis of DR.

Objectives: To systematically review the existing literature evaluating the diagnostic performance, feasibility, and implementation of AI-based models for early detection of diabetic retinopathy.

Methods: A systematic search was conducted across PubMed, Scopus, Web of Science, Embase, and IEEE Xplore for peer-reviewed studies published between 2010 and May 2024. Eligibility criteria included adult diabetic populations screened using AI tools, with outcomes reported in terms of sensitivity, specificity, or AUC. Results: Fifteen studies met the inclusion criteria. CNN-based models demonstrated high diagnostic accuracy, with sensitivity ranging from 87.2% to 94.1% and specificity between 90.7% and 98.5%. Explainable AI (XAI) improved clinician trust. Limitations included variability in datasets and lack of longitudinal clinical impact assessment.

Conclusion: AI, especially deep learning models, shows strong potential to enhance DR screening. However, ethical implementation, population-specific validation, and longitudinal effectiveness remain areas requiring further research.